Character Encodings and stringi
This man page aims to explain (or at least to cast light on) how stringi deals with character strings in various encodings. In particular you should note that:
- R lets strings in ASCII, UTF-8, and your platform's
native encoding coexist peacefully. Character vector output
catetc. silently reencodes each string so that it can be properly shown e.g. in the R's console.
- Functions in stringi process each string internally in Unicode, which is a superset of all character representation schemes. Even if a string is given in the native encoding, i.e. your platform's default one, it will be converted to Unicode (precisely: UTF-8 or UTF-16).
- Most stringi functions always return UTF-8 encoded strings, regardless of the input encoding. What is more, the functions have been optimized for UTF-8/ASCII input (they have competitive, if not better performance, especially when doing more complex operations like string comparison, sorting, and even concatenation). Thus, it is best to rely on cascading calls to stringi operations solely.
"Hundreds of encodings have been developed over the years, each for small
groups of languages and for special purposes. As a result,
the interpretation of text, input, sorting, display, and storage
depends on the knowledge of all the different types of character sets
and their encodings. Programs have been written to handle either
one single encoding at a time and switch between them, or to convert
between external and internal encodings."
"Unicode provides a single character set that covers the major languages of the world, and a small number of machine-friendly encoding forms and schemes to fit the needs of existing applications and protocols. It is designed for best interoperability with both ASCII and ISO-8859-1 (the most widely used character sets) to make it easier for Unicode to be used in almost all applications and protocols" (see the ICU User Guide).
The Unicode Standard determines the way to map any possible character to a numeric value -- a so-called code point. Such code points, however, have to be stored somehow in computer's memory. The Unicode Standard encodes characters in the range U+0000..U+10FFFF, which amounts to a 21-bit code space. Depending on the encoding form (UTF-8, UTF-16, or UTF-32), each character will then be represented either as a sequence of one to four 8-bit bytes, one or two 16-bit code units, or a single 32-bit integer (compare the ICU FAQ).
In most cases, Unicode is a superset of the characters supported by any given code page.
UTF-8 and UTF-16
For portability reasons, the UTF-8 encoding is the most natural choice
for representing Unicode character strings in R. UTF-8 has ASCII as its
subset (code points 1--127 represent the same characters in both of them).
Code points larger than 127 are represented by multi-byte sequences
(from 2 to 4 bytes: by the way, not all sequences of bytes are valid UTF-8,
stri_enc_isutf8). Most of the computations in stringi are performed internally
using either UTF-8 or UTF-16 encodings (this depends on type of service
you request: some ICU services are designed only to work with UTF-16).
Thanks to such a choice, with stringi you get the same result on
each platform, which is -- unfortunately -- not the case of base R's
functions (it is for example known that performing a regular expression
search under Linux on some texts may give you a different result
to those obtained under Windows). We really had portability in our minds
while developing our package! We have observed that R correctly handles UTF-8 strings regardless of your
platform's native encoding (see below). Therefore, we decided that most
functions in stringi will output its results in UTF-8
-- this speeds ups computations on cascading calls to our functions:
the strings does not have to be re-encoded each time. Note that some Unicode characters may have an ambiguous representation.
For example, ``a with ogonek'' (one character) and ``a''+``ogonek''
(two graphemes) are semantically the same. stringi provides functions
to normalize character sequences, see
for discussion. However, it is observed that denormalized strings
do appear very rarely in typical string processing activities. Additionally, do note that stringi silently removes byte order marks
(BOMs - they may incidentally appear in a string read from a text file)
from UTF8-encoded strings, see
Character Encodings in R
You should keep in mind that data in memory are just bytes (small integer
values) -- an encoding is a way to represent characters with such
numbers, it is a semantic "key" to understand a given byte sequence.
For example, in ISO-8859-2 (Central European), the value 177 represents
Polish ``a with ogonek'', and in ISO-8859-1 (Western European),
the same value denotes the ``plus-minus'' sign. Thus, a character encoding
is a translation scheme: we need to communicate with R somehow,
relying on how it represents strings. Basically, R has a very simple encoding marking mechanism,
stri_enc_mark. There is an implicit assumption
that your platform's default (native) encoding is always a superset
of ASCII -- stringi checks that when your native encoding
is being detected automatically on ICU's initialization and each time
when you change it manually by calling
stri_enc_set. Character strings in R (internally) can be declared to be in:
latin1, i.e. ISO-8859-1 (Western European);
bytes-- for strings that should be manipulated as sequences of bytes.
- ASCII -- for strings consisting only of byte codes not greater than 127;-
Encoding; quite a misleading name: no explicit encoding mark) -- for strings that are assumed to be in your platform's native (default) encoding. This can represent UTF-8 if you are an OS X user, or some 8-bit Windows code page, for example. The native encoding used by R may be determined by examining the LC_CTYPE category, see
stri_enc_get(unless you know what you are doing, the default encoding should only be changed if the automatic encoding detection process fails on stringi load). Functions which allow
"bytes"encoding markings are very rare in stringi, and were carefully selected. These are:
stri_encode. Finally, note that R lets strings in ASCII, UTF-8, and your platform's native encoding coexist peacefully. Character vector printed with
catetc. silently reencodes each string so that it can be properly shown e.g. on the console.
Apart from automatic conversion from the native encoding,
you may re-encode a string manually, for example
when you load it from a file saved in a different platform.
stri_enc_list for the list of
encodings supported by ICU.
Note that converter names are case-insensitive
and ICU tries to normalize the encoding specifiers.
Leading zeroes are ignored in sequences of digits (if further digits follow),
and all non-alphanumeric characters are ignored. Thus the strings
"UTF-8", "utf_8", "u*Tf08" and "Utf 8" are equivalent. The
allows you to convert between any given encodings
(in some cases you will obtain
strings, or even lists of raw vectors (i.e. for UTF-16).
There are also some useful more specialized functions,
stri_enc_toutf32 (converts a character vector to a list
of integers, where one code point is exactly one numeric value)
stri_enc_toascii (substitutes all non-ASCII
bytes with the SUBSTITUTE CHARACTER,
which plays a similar role as R's
NA value). There are also some routines for automated encoding detection,
Given a text file, one has to know how to interpret (encode)
raw data in order to obtain meaningful information. Encoding detection is always an imprecise operation and
needs a considerable amount of data. However, in case of some
encodings (like UTF-8, ASCII, or UTF-32) a ``false positive'' byte
sequence is quite rare (statistically speaking). Check out
stri_enc_detect (among others) for a useful
function in this category.
Unicode Basics -- ICU User Guide, http://userguide.icu-project.org/unicode Conversion -- ICU User Guide, http://userguide.icu-project.org/conversion Converters -- ICU User Guide, http://userguide.icu-project.org/conversion/converters (technical details) UTF-8, UTF-16, UTF-32 & BOM -- ICU FAQ, http://www.unicode.org/faq/utf_bom.html